Storage System IO Access Prediction Using Time-Segmented Tendency Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current storage systems inaccurately predict IO access concentration durations due to changes in distribution over time, leading to ineffective data migration strategies between HDD and SSD in tiered storage systems.
Innovation Solution
An information processing apparatus that generates data access prediction information by comparing IO access concentration tendencies before and after a standard deviation change, regenerating predictions to exclude outdated data access history and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the total duration time of IO access concentration is used for prediction over a long span (e.g., one day), then the prediction covers a broad time range, but the accuracy decreases when the distribution of duration time changes during the day
Solution Approach 1:
The patent divides the prediction time span into multiple segments (first time period and second time period) rather than using a single long-span total. By segmenting the time range and comparing tendency information from different segments, the system can detect changes in access patterns and regenerate predictions with appropriate time spans, thus maintaining both broad coverage and high accuracy.
Solution Approach 2:
The patent implements a dynamic prediction system that adapts to changing access patterns. By continuously comparing concentration state tendency information from different time periods and detecting variations beyond a standard value, the system dynamically adjusts the prediction time span and regenerates predictions when necessary, making the prediction mechanism flexible and adaptive to temporal changes.
2Quantity of substance
If data access history from both before and after the change is included in the total, then the sample size increases, but the prediction accuracy decreases due to mixing different distribution patterns
Solution Approach 1:
The patent extracts and separates data from different time periods with different access patterns. By comparing tendency information from the first and second time periods and identifying when they differ significantly, the system extracts only the relevant data (from the current time period) for prediction, excluding outdated data with different distributions. This ensures high prediction accuracy by using homogeneous data with consistent access patterns.
3Device complexity
If the prediction system uses a fixed time span, then the system is simple to implement, but it cannot adapt to changing IO access concentration patterns
Solution Approach 1:
The patent transforms the fixed-time-span prediction system into a dynamic one by introducing comparison mechanisms between different time periods and regeneration logic. The system automatically adjusts the effective prediction time span based on detected changes in access patterns, maintaining simplicity in implementation while achieving adaptability through automated temporal analysis and selective data usage.
Data Source
AI summary
An IO access concentration is precisely predicted by comparing first concentration state tendency information and second concentration state tendency information indicating a tendency of a concentration state of a unit region being in the concentration state, the first concentration state tendency information being collected during a first time period, the second concentration state tendency information being collected during a second time period prior to the first time period, and when the first concentration state tendency information is detected to vary by a standard value or more from the second concentration state tendency information, regenerating a data access prediction information by excluding a data access history collected during the second time period.